|Year : 2018 | Volume
| Issue : 4 | Page : 65-72
Validation of a modified pediatric risk of mortality III model in a pediatric intensive care unit in Thailand
Kanokpan Ruangnapa1, Sittikiat Sucheewakul1, Tippawan Liabsuetrakul2, Edward McNeil2, Kantara Lim1, Wanaporn Anantaseree1
1 Department of Pediatrics, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla, Thailand
2 Epidemiology Unit, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla, Thailand
|Date of Web Publication||28-Dec-2018|
Department of Pediatrics, Faculty of Medicine, Prince of Songkla University, Hat Yai, Songkhla 90110
Source of Support: None, Conflict of Interest: None
Objective: The objective of this study is to compare the performance of a modified Pediatric Risk of Mortality (PRISM) III model with the original PRISM III in prediction of mortality risk in a Thailand pediatric intensive care unit (PICU). Subjects and Methods: Children aged 1 month to 18 years who stayed in the PICU for more than 8 h during November 2013 to December 2016 were included in the study. Results: The medical records of 1175 PICU patients were included in the analysis. The patients were randomly split into two equal groups: a development (n = 588) and a validation (n = 587) sample. A modified PRISM III model was derived from the original PRISM III by omitting arterial blood gas parameters and adding selected clinical variables. The model was developed using a multiple logistic regression model on the development sample and assessed using the area under the curve (AUC) obtained from a receiver operating characteristic curve. The modified PRISM III scores were significantly higher in nonsurvivors (median = 9, interquartile range [IQR] = 4 − 13) compared to survivors (median = 2, IQR = 0 − 5). The modified PRISM III model had similar discriminative performances compared to the original PRISM III in predicting 2-day mortality (AUC: 0.874 vs. 0.873), 7-day mortality (AUC: 0.851 vs. 0.851) and overall mortality (AUC: 0.845 vs. 0.956). The modified PRISM III model was calibrated in the validation sample, and the standardized mortality ratios (SMRs) were similar. Conclusions: The performance of a modified PRISM III model in predicting mortality risk was comparable to the original PRISM III. Both had similar discriminative performance and SMR for overall mortality prediction in a PICU.
Keywords: Mortality prediction, pediatric intensive care unit, Pediatric Risk of Mortality III
|How to cite this article:|
Ruangnapa K, Sucheewakul S, Liabsuetrakul T, McNeil E, Lim K, Anantaseree W. Validation of a modified pediatric risk of mortality III model in a pediatric intensive care unit in Thailand. Pediatr Respirol Crit Care Med 2018;2:65-72
|How to cite this URL:|
Ruangnapa K, Sucheewakul S, Liabsuetrakul T, McNeil E, Lim K, Anantaseree W. Validation of a modified pediatric risk of mortality III model in a pediatric intensive care unit in Thailand. Pediatr Respirol Crit Care Med [serial online] 2018 [cited 2020 Nov 27];2:65-72. Available from: https://www.prccm.org/text.asp?2018/2/4/65/249000
| Introduction|| |
Critically ill children admitted in a pediatric intensive care unit (PICU) have a high risk of mortality. Various severity scoring systems have been developed to classify PICU patients on admission and determine their risk of mortality. The Pediatric Risk of Mortality (PRISM) III is a widely accepted scoring system, which uses the worst physiologic and laboratory values of patients on the day of admission to predict PICU mortality, similar to other scoring systems.,
PICU patients who die earlier are usually patients who had more severe illness on admission and often die within 48 h after admission., Severity scoring systems such as PRISM III, which assesses patients on admission, may perform better for mortality prediction in these patients. However, advances in critical care have resulted in a lengthening of PICU admissions and an increase in the number of long-stay patients. These patients utilize more hospital resources and require more intensive therapy such as mechanical ventilation., In addition, the discriminative performance of various severity scoring systems has shown poorer performance in prediction of mortality among long-stay PICU patients.,
From a previous study, the discriminative performance of PRISM III for mortality prediction was improved by adding eight physiologic variables measured during admission to the PICU. As mentioned above, the dynamic conditions of the patient and management in the PICU can alter the mortality risk, especially in long-stay patients. Critically ill patients who require intense or rescue therapy such as mechanical ventilation,, high-frequency oscillatory ventilation (HFOV),, or cardiopulmonary resuscitation have an increased risk of mortality. To achieve a good prediction model, both discrimination and calibration assessment are essential.
Another limitation of PRISM III is the requirement for arterial blood gas measurements, which is arguably unethical in children who have no prior clinical indications., Due to the nonfeasibility of performing invasive arterial blood gas measurements in some children and the benefit of adding other clinical variables, we modified the original PRISM III by removed the blood gas parameters and added a number of clinical variables we felt would create a better prognostic picture. This study aimed to assess the performance of the modified PRISM III model compared to the original PRISM III in predicting mortality.
| Subjects and Methods|| |
A retrospective study was conducted among children aged 1 month–18 years at the PICU of Prince of Songkla University, Songkhla, Thailand. The unit has eight beds with provisions for a mechanical ventilator and continuous hemodynamic monitoring. The study was approved by the Ethics Committee of the Faculty of Medicine, Prince of Songkla University.
The medical records of patients admitted to the PICU during November 2013–December 2016 were evaluated. Patients who were re-admitted during the study period were counted separately. Patients who had <8 h stay and were admitted for postprocedural care, including postcardiac catheterization or diagnostic bronchoscopy, were excluded. The required sample size was calculated based on a type I error of 5%, power of 90%, an overall mortality rate of 12.0%, and a sensitivity of 0.85. Using the formula for two-independent proportions, at least 1030 participants were required for validating the mortality prediction model.
Eligible patients were identified from the registration database of the PICU. All required PRISM III variables were recorded using the most abnormal values of physiologic and laboratory data within 24 h of admission to the PICU. The scoring method of the original PRISM III was followed. A PRISM III score ranges from 0 to 74 and consists of five physiologic and 12 laboratory variables as described by Pollack et al. which are categorized into four groups: cardiovascular/neurologic vital signs (range: 0–30), acid-based/blood gasses (range: 0–22), biochemical tests (range: 0–10), and hematological tests (range: 0–12). The modified PRISM III scores (range: 0–52) was derived from the PRISM III scores by excluding the scores of all acid-based/blood gas variables [Table 1].
|Table 1: Scores for the modified Pediatric Risk of Mortality III by omitting arterial blood gas measurements from original PRISM III|
Click here to view
Apart from the variables measured in PRISM III, selected patient demographic characteristics and clinical data were additionally collected and analyzed. These variables were selected based on scientific logic and previous studies investigating risk factors for mortality among patients admitted to an PICU.,, There were two groups of variables. The first group, called “admission variables,” included various factors from the day of PICU admission, notably postoperative care, postcardiac surgery, nonoperative cardiovascular diseases, chromosomal anomaly, malignancy, previous admission to a PICU, cardiopulmonary resuscitation (CPR) before admission, acute diabetic ketoacidosis, intubation on admission, and serum albumin level. Because the clinical condition of PICU patients is so dynamic, the second group of variables we called “therapeutic variables” that occurred during the course of the PICU stay was recorded for the overall mortality prediction model, namely mechanical ventilation use, HFOV use, inhaled nitric oxide use, and renal replacement therapy. All these variables except albumin level were coded as yes or no. All data were reviewed and verified by nurses who were not involved in the patient's care.
The data were analyzed with R software version 3.4.0. The patients were randomly split into two equal groups, a development sample and a validation sample. The data were summarized using means with standard variations and medians with interquartile ranges (IQRs) as appropriate. The characteristics and clinical variables between the survivor and nonsurvivor groups were compared with the Chi-squared and Fisher's exact tests for categorical variables and Student's t-test or Mann–Whitney test for continuous variables. P < 0.05 was considered statistically significant.
All variables with P < 0.2 from the univariate analysis were included into the initial multivariate logistic regression model predicting death. For 2- and 7-day mortality, the original PRISM III scores were compared with the modified PRISM III scores with the additional admission variables. For overall mortality, the original PRISM III scores were compared with the modified PRISM III scores with the addition of the admission and therapeutic variables. The best model was selected based on the lowest Akaike's information criterion value. The discrimination capacity between nonsurvivors and survivors of the best model was assessed using receiver operating characteristic curves based on the area under the curve (AUC). An AUC between 0.80 and 0.90 represents good discrimination whereas an AUC ≥0.90 represents excellent discrimination. Calibration of models was assessed in the validation sample using the standardized mortality ratio (SMR) calculated by dividing the observed number of deaths with the expected number within each decile of probability of mortality.
| Results|| |
There were 1202 PICU admissions during the study period, of which 27 were excluded because the length of PICU stay was less than 8 h (n = 19) or the reason for admission was diagnostic bronchoscopy (n = 8). A total of 1175 admissions were, therefore, included in the analysis, which were randomly split into the development (n = 588) and validation (n = 587) samples.
For the development sample, 53% were male and 58% were aged <5 years. The most common reasons for admission to PICU were cardiovascular (36.5%) and respiratory (30.1%) problems followed by neurological problems (19.1%). Fifty-two percent of the patients were admitted for treatment of medical illnesses and 48% for postoperative care. Among the postoperative care patients, 22% underwent emergency/unscheduled surgery and 47.7% had cardiac surgery, including both open and closed heart surgeries. The median length of PICU stay was 3.5 days (IQR: 2−7.2), and the PICU mortality rate was 13.9%. A comparison of the patients' demographic characteristics and clinical variables between survivors and nonsurvivors is shown in [Table 2]. Significantly higher modified PRISM III scores were found for nonsurvivors (median [IQR] = 9 [4−13]), compared to survivors (median [IQR] = 2 [0−5]) and PRISM III scores (median [IQRs] = 12 [8−18] vs. 4 [0−7]), respectively.
[Table 3] shows the factors significantly associated with overall mortality from the multiple logistic regression model. Four admission variables, such as postoperative care, CPR before admission, intubation on admission, and serum albumin level, and three therapeutic variables, such as mechanical ventilation use, HFOV use, and CPR during PICU stay, were significantly associated with overall mortality. These seven variables were used in the overall mortality prediction model.
|Table 3: Significant variables associated with mortality from multiple logistic regression|
Click here to view
The discriminative performances of the modified PRISM III with admission variables and original PRISM III for 2-day and 7-day mortality are shown in [Figure 1]. The modified PRISM III was comparable to the original PRISM III in predicting both 2-day (AUC 0.874 vs. 0.873) and 7-day mortalities (AUC 0.851 vs. 0.853), of which the mortality prediction was better for 2-day than for 7-day mortality.
|Figure 1: Receiver operating characteristic curves for (a) 2-day mortality and (b) 7-day mortality prediction using the modified Pediatric Risk of Mortality III score with admission variables versus the original Pediatric Risk of Mortality III. The additional admission variables were postoperative care, cardiopulmonary resuscitation before admission, intubation on admission, and serum albumin. PRISM III: Pediatric Risk of Mortality III, AUC: Area under the curve.|
Click here to view
The prediction models for overall mortality are illustrated in [Figure 2]. The modified PRISM III with combined variables was found to give the best prediction. Based on the AUC, the discriminative performance of this modified PRISM III model (AUC = 0.956) was better than the modified PRISM III with admission variables alone (AUC = 0.850) and the original PRISM III (AUC = 0.850).
|Figure 2: Receiver operating characteristic curves for overall mortality prediction using the different models. The combined variables consist of four admission variables (postoperative care, cardiopulmonary resuscitation before admission, intubation on admission, and serum albumin) and three therapeutic variables (mechanical ventilation use, high-frequency oscillatory ventilation use, and cardiopulmonary resuscitation during pediatric intensive care unit stay). PRISM III: Pediatric Risk of Mortality III, AUC: Area under the curve.|
Click here to view
In the validation sample, the PICU mortality rate was 13.6%, which was similar to the developmental sample. The observed and expected mortality rates for the validation samples using the modified PRISM III with admission and therapeutic variables and original PRISM III are shown in [Table 4]. For all deciles of mortality probability, the SMR was close to one except for probabilities in the range of 0–0.1 which gave an SMR of 0.23 indicating an overprediction of mortality. The average standardized mortality rates from the modified PRISM III and original PRISM III were 0.99 and 0.98, respectively.
|Table 4: Calibration of the modified Pediatric Risk of Mortality III model in a validation sample (n=587; 80 deaths) to predict overall mortality|
Click here to view
| Discussion|| |
A modified PRISM III model which omitted arterial blood gas variables and added some clinical variables had a similar discriminative performance compared to the original PRISM III. The addition of admission variables was useful in the modified PRISM III model in predicting 2-day (AUC: 0.874 vs. 0.873) and 7-day mortality (AUC: 0.851 vs. 0.853), and the addition of therapeutic variables was useful for predicting overall mortality (AUC: 0.956 vs. 0.845). These modified PRISM III models can be applied when blood gas measurements are not feasible.
In modernized pediatric critical care, there is an increasing trend to use noninvasive techniques where feasible. For example, previous studies have suggested the use of oxygen saturation (SpO2) obtained by pulse oximetry instead of arterial partial pressure of oxygen (PaO2) in respiratory distress, even among mechanically ventilated children, and encouraged the use of end-tidal CO2 to estimate PaCO2 to assess the severity of disease and follow treatment response. The use of noninvasive ventilation and high-flow nasal cannulae as initial respiratory support in various types of cardiorespiratory failure in children, including those with postoperative conditions, has been increasing;,, thus, arterial blood gas monitoring is being performed less frequently and might even be considered unethical in this population. A recent study by Ray et al. demonstrated the feasibility of using SpO2/FiO2 instead of PaO2/FiO2 values for calculating pediatric index of mortality (PIM) scores for use in predicting PICU mortality.
Our mortality prediction model demonstrated that the addition of important admission and therapeutic variables could improve the discriminative ability of the original PRISM III model. Due to dynamic changes in any critically ill patient's clinical status, the variables from the first few hours on the day of admission may not be sufficient to reflect the condition of long-stay patients with the PRISM III. Visser et al. demonstrated that both the PRISM and PIM models have poorer discriminative performance for patients who stay in a PICU longer than 6 days compared to those who stay ≤6 days. The median length of stay for nonsurvivors in our study was 6 days, which was similar to two other studies., This is the reason that our modified PRISM, with the addition of therapeutic variables during PICU stay, seems to be more useful in mortality prediction.
Additional variables used in the overall mortality prediction model of our study included postoperative care, serum albumin level, intubation on admission, CPR before admission, mechanical ventilator use, HFOV use, and CPR during the PICU stay. Some of these variables were associated with mortality in previous studies among critically ill patients.,,,, In our study, postoperative care patients had a lower mortality risk compared to those with medical illnesses. Approximately 80% of postoperative care patients admitted to a PICU have undergone elective surgery while medically ill patients who are admitted to a PICU usually have complicated problems leading to multiorgan dysfunction, which is known as a strong predictor of mortality., Furthermore, serum albumin on admission, another independent predictor of mortality in our study, has been reported to be associated with higher mortality and poor PICU outcome including requiring prolonged mechanical ventilator use, longer duration of PICU stay, and risk of progression to multiorgan dysfunction., Tiwari et al. reported that hypoalbuminemia at admission was associated with a higher 60-day mortality and lower probability of discharge from intensive care, and an increase of 1.0 g/dL in serum albumin at admission resulted in a 73% reduction in mortality the hazard of death. Receiving CPR, either before PICU admission or during the course of their PICU stay, has also been reported as an independent factor associated with predicting mortality. Although the success rate (return to spontaneous circulation) of in-hospital pediatric CPR has been reported to be as high as 60%–75%,, both studies found that survival rate to discharge was only 20%. Another study reported that children who survived from CPR usually suffered from multiorgan dysfunction and metabolic disturbances and had high susceptibility to infection and metabolic disturbances and factors which are all associated with mortality.
Cardiorespiratory failure and the need for respiratory support were the most common reasons for PICU admission in our study. As currently recommended, our practice has been gradually changing to the use of noninvasive ventilation as first-line respiratory support in order to reduce PICU stay, ventilator-associated pneumonia, and postextubation complications. Therefore, the requirement of intubation on admission as well as mechanical ventilation use would reflect failure of noninvasive ventilation or higher severity of illness and can be used as a predictor for mortality, similar to previous studies., Yaman et al. reported that the PICU patients who had failure of noninvasive ventilation use had higher PRISM III scores, more frequent underlying disease, longer PICU stay, and increased risk of mortality. Likewise, HFOV is often used as rescue therapy, particularly in patients with acute respiratory distress syndrome (ARDS) with refractory hypoxemia. Currently, there is insufficient evidence to support the benefit of HFOV use in terms of mortality reduction., Furthermore, the use of HFOV increases the use of sedatives and vasoactive and neuromuscular blockage agents. HFOV has been reported to be associated with poor PICU outcomes and mortality. In our practice, we do not use the early HFOV strategy, so patients receiving HFOV reflect the presence of moderate to severe ARDS and an unresponsiveness to conventional ventilation, a combination of factors which has been reported to be associated with a high risk of mortality.
For the probability of mortality risk at all deciles, the SMR showed good concordance, except at the deciles of p 0–0.1. These findings are consistent with previous studies using the PRISM III model that also found good predictability of SMR for higher probabilities of mortality exceeding 0.5, although different scales of stratification were used.,,, The overprediction of SMR at deciles of P 0–0.1 in our study may have been due to a lower number of less severe patients in our study resulting in a large proportion of survivors in this decile. Slater et al. observed a significant overprediction of deaths in the subgroups of respiratory (SMR = 0.65) and cardiac patients (SMR = 0.54), which constituted nearly two-thirds of our study population.
There were some limitations in our study. First, the study had a small sample size compared to the original PRISM III study. However, our patient sample was adequate based on our sample size calculation. Second, all the datasets we used were from a single PICU. However, the parameters applied in our study are common practice for PICU patients in most countries in Asia.,,
| Conclusions|| |
The modified PRISM III model, omitting arterial blood gas parameters with additional clinical and therapeutic variables, gave a similar discriminative performance for overall mortality prediction in PICU to the original PRISM III. The modified PRISM III model can be applied when blood gas measurements are not feasible. However, the risk scoring system using variables in our modified PRISM III model should be further validated in large-scale populations.
This study was granted by the Faculty of Medicine, Prince of Songkla University. We would like to thank all staff of the Research Analysis and Publication: Intensive Development (RAPID) project, Epidemiology Unit, Faculty of Medicine, Prince of Songkla University, for their assistance with data analysis and manuscript preparation.
Financial support and sponsorship
Conflicts of interest
There are no conflicts of interest.
| References|| |
Marcin JP, Pollack MM. Review of the acuity scoring systems for the pediatric Intensive Care Unit and their use in quality improvement. J Intensive Care Med 2007;22:131-40.
Pollack MM, Patel KM, Ruttimann UE. PRISM III: An updated pediatric risk of mortality score. Crit Care Med 1996;24:743-52.
Brady AR, Harrison D, Black S, Jones S, Rowan K, Pearson G, et al.
Assessment and optimization of mortality prediction tools for admissions to pediatric intensive care in the United Kingdom. Pediatrics 2006;117:e733-42.
Gonçalves JP, Severo M, Rocha C, Jardim J, Mota T, Ribeiro A, et al.
Performance of PRISM III and PELOD-2 scores in a pediatric Intensive Care Unit. Eur J Pediatr 2015;174:1305-10.
Khajeh A, Noori NM, Reisi M, Fayyazi A, Mohammadi M, Miri-Aliabad G, et al.
Mortality risk prediction by application of pediatric risk of mortality scoring system in pediatric Intensive Care Unit. Iran J Pediatr 2013;23:546-50.
El-Nawawy A. Evaluation of the outcome of patients admitted to the pediatric Intensive Care Unit in Alexandria using the pediatric risk of mortality (PRISM) score. J Trop Pediatr 2003;49:109-14.
Briassoulis G, Filippou O, Natsi L, Mavrikiou M, Hatzis T. Acute and chronic paediatric intensive care patients: Current trends and perspectives on resource utilization. QJM 2004;97:507-18.
Marcin JP, Slonim AD, Pollack MM, Ruttimann UE. Long-stay patients in the pediatric Intensive Care Unit. Crit Care Med 2001;29:652-7.
Visser IH, Hazelzet JA, Albers MJ, Verlaat CW, Hogenbirk K, van Woensel JB, et al.
Mortality prediction models for pediatric intensive care: Comparison of overall and subgroup specific performance. Intensive Care Med 2013;39:942-50.
Costa GA, Delgado AF, Ferraro A, Okay TS. Application of the pediatric risk of mortality (PRISM) score and determination of mortality risk factors in a tertiary pediatric Intensive Care Unit. Clinics (Sao Paulo) 2010;65:1087-92.
Faraci M, Bagnasco F, Giardino S, Conte M, Micalizzi C, Castagnola E, et al.
Intensive Care Unit admission in children with malignant or nonmalignant disease: Incidence, outcome, and prognostic factors: A single-center experience. J Pediatr Hematol Oncol 2014;36:e403-9.
Gupta P, Green JW, Tang X, Gall CM, Gossett JM, Rice TB, et al.
Comparison of high-frequency oscillatory ventilation and conventional mechanical ventilation in pediatric respiratory failure. JAMA Pediatr 2014;168:243-9.
Rettig JS, Smallwood CD, Walsh BK, Rimensberger PC, Bachman TE, Bollen CW, et al.
High-frequency oscillatory ventilation in pediatric acute lung injury: A multicenter international experience. Crit Care Med 2015;43:2660-7.
Parra DA, Totapally BR, Zahn E, Jacobs J, Aldousany A, Burke RP, et al.
Outcome of cardiopulmonary resuscitation in a pediatric cardiac Intensive Care Unit. Crit Care Med 2000;28:3296-300.
Steyerberg EW, Vickers AJ, Cook NR, Gerds T, Gonen M, Obuchowski N, et al.
Assessing the performance of prediction models: A framework for traditional and novel measures. Epidemiology 2010;21:128-38.
Khemani RG, Thomas NJ, Venkatachalam V, Scimeme JP, Berutti T, Schneider JB, et al.
Comparison of SpO2 to PaO2 based markers of lung disease severity for children with acute lung injury. Crit Care Med 2012;40:1309-16.
Baudin F, Bourgoin P, Brossier D, Essouri S, Emeriaud G, Wysocki M, et al.
Noninvasive estimation of arterial CO2 from end-tidal CO2 in mechanically ventilated children: The GRAeDIENT pilot study. Pediatr Crit Care Med 2016;17:1117-23.
Ngamjarus C, Chongsuvivatwong V. N4Studies: Sample Size and Power Calculations for iOS. The Royal Golden Jubilee Ph.D. Program – The Thailand Research Fund & Prince of Songkla University; 2014.
R Core Team. R: A Language and Environment for Statistical Computing. Foundation for Statistical Computing. Vienna, Austria; 2017. Available from: https://www.R-project.org/
. [Last accessed on 2018 Feb 19].
Zweig MH, Campbell G. Receiver-operating characteristic (ROC) plots: A fundamental evaluation tool in clinical medicine. Clin Chem 1993;39:561-77.
Yaman A, Kendirli T, Ödek Ç, Ateş C, Taşyapar N, Güneş M, et al.
Efficacy of noninvasive mechanical ventilation in prevention of intubation and reintubation in the pediatric Intensive Care Unit. J Crit Care 2016;32:175-81.
Coletti KD, Bagdure DN, Walker LK, Remy KE, Custer JW. High-flow nasal cannula utilization in pediatric critical care. Respir Care 2017;62:1023-9.
Kneyber MC, de Luca D, Calderini E, Jarreau PH, Javouhey E, Lopez-Herce J, et al.
Recommendations for mechanical ventilation of critically ill children from the paediatric mechanical ventilation consensus conference (PEMVECC). Intensive Care Med 2017;43:1764-80.
Ray S, Rogers L, Pagel C, Raman S, Peters MJ, Ramnarayan P, et al.
PaO2/FIO2 ratio derived from the spO2/FIO2 ratio to improve mortality prediction using the pediatric index of mortality-3 score in transported intensive care admissions. Pediatr Crit Care Med 2017;18:e131-6.
Leite HP, Rodrigues da Silva AV, de Oliveira Iglesias SB, Koch Nogueira PC. Serum albumin is an independent predictor of clinical outcomes in critically ill children. Pediatr Crit Care Med 2016;17:e50-7.
Tiwari LK, Singhi S, Jayashree M, Baranwal AK, Bansal A. Hypoalbuminemia in critically sick children. Indian J Crit Care Med 2014;18:565-9.
] [Full text]
Wu ET, Li MJ, Huang SC, Wang CC, Liu YP, Lu FL, et al.
Survey of outcome of CPR in pediatric in-hospital cardiac arrest in a medical center in Taiwan. Resuscitation 2009;80:443-8.
Sutton RM, Morgan RW, Kilbaugh TJ, Nadkarni VM, Berg RA. Cardiopulmonary resuscitation in pediatric and cardiac Intensive Care Units. Pediatr Clin North Am 2017;64:961-72.
Morris JV, Ramnarayan P, Parslow RC, Fleming SJ. Outcomes for children receiving noninvasive ventilation as the first-line mode of mechanical ventilation at intensive care admission: A Propensity score-matched cohort study. Crit Care Med 2017;45:1045-53.
Qiao JY, Li YZ, Wang HY, Zhang SD. A meta analysis of the efficacy of high-frequency oscillatory ventilation versus conventional mechanical ventilation for treating pediatric acute respiratory distress syndrome. Zhongguo Dang Dai Er Ke Za Zhi 2017;19:430-5.
Sud S, Sud M, Friedrich JO, Wunsch H, Meade MO, Ferguson ND, et al.
High-frequency oscillatory ventilation versus conventional ventilation for acute respiratory distress syndrome. Cochrane Database Syst Rev 2016;4:CD004085.
Ferguson ND, Cook DJ, Guyatt GH, Mehta S, Hand L, Austin P, et al.
High-frequency oscillation in early acute respiratory distress syndrome. N Engl J Med 2013;368:795-805.
Wong JJ, Phan HP, Phumeetham S, Ong JS, Chor YK, Qian S, et al.
Risk stratification in pediatric acute respiratory distress syndrome: A multicenter observational study. Crit Care Med 2017;45:1820-8.
Slater A, Shann F; ANZICS Paediatric Study Group. The suitability of the pediatric index of mortality (PIM), PIM2, the pediatric risk of mortality (PRISM), and PRISM III for monitoring the quality of pediatric intensive care in Australia and New Zealand. Pediatr Crit Care Med 2004;5:447-54.
Choi KM, Ng DK, Wong SF, Kwok KL, Chow PY, Chan CH, et al.
Assessment of the pediatric index of mortality (PIM) and the pediatric risk of mortality (PRISM) III score for prediction of mortality in a paediatric Intensive Care Unit in Hong Kong. Hong Kong Med J 2005;11:97-103.
Santhanam H, Ong JS, Shen L, Tan PL, Koh PL. Risk factors for mortality in Asian children admitted to the paediatric Intensive Care Unit after haematopoietic stem cell transplantation. Ann Acad Med Singapore 2017;46:44-9.
Yang X, Qian SY, Zhu YM, Xu X, Liu CF, Xu F, et al.
Survey on the prevalence of continuous blood purification in Chinese pediatric critical care. Zhonghua Er Ke Za Zhi 2018;56:128-33.
[Figure 1], [Figure 2]
[Table 1], [Table 2], [Table 3], [Table 4]